AMD: Big talk, small results — When will it execute?
AMD (AMD.O) released its Q4 2025 results (quarter ended Dec 2025) after U.S. market close on the morning of Feb 4, 2026 Beijing time. Key takeaways:
1) Headline results: Revenue came in at $10.27bn, +34.1% YoY, beating the raised consensus (~$10.0bn). Street models had moved up recently on a rebound in server CPU demand. Growth this quarter was driven primarily by data center CPUs and the Client segment.
$AMD(AMD.US) GAAP GPM was 54.3%, up 360bps YoY, helped by the sell-through of MI308 inventory (~$360mn). Ex-MI308, underlying GPM was ~50.8%, roughly flat YoY.
2) Opex: R&D expense was $2.33bn (+36.1% YoY); SG&A was $1.20bn (+51% YoY). With topline growing fast, core opex also climbed, taking the core opex ratio to 34.4%, up 170bps YoY.
Net income was $1.5bn, largely impacted by non-recurring items. On an operating basis, core OP was $2.05bn with a 20% OPM, continuing to improve.
3) Segment mix: Data Center and Client together accounted for over 70% of revenue.
1) Client share gains: Segment revenue rose to $3.1bn, +34% YoY. Global PC shipments grew 11% YoY, while AMD’s Client revenue rose much faster, reflecting continued share gains in PCs.
2) Data Center awaiting new product lift: Segment revenue was $5.38bn, +39% YoY. Growth was driven by server CPUs and increased volume of the MI355 family.
i) Server GPU: soft again. This quarter included $390mn of MI308 sales to China, which are booked under server GPUs.
Since last quarter did not include MI308 revenue, ex-MI308, we estimate server GPU revenue at ~$2.0bn this quarter, up only ~$150mn QoQ, which is underwhelming.
While the market is more focused on the MI450 family ramp in 2H26, AMD’s first shift from single-die to rack-scale cluster delivery, the MI355’s slowing QoQ trajectory still weighs on sentiment.
ii) Server CPU: still strong. The main incremental driver this quarter was share gains in server CPUs. Third-party data indicate AMD’s server CPU unit share has risen above 20%.
4) Guidance: Q1 2026 revenue of $9.5–10.1bn (vs. raised consensus ~$9.7bn), with the midpoint ($9.8bn) implying -4.5% QoQ, and includes ~$100mn of MI308 sales to China. Non-GAAP GPM guided to ~55% (vs. consensus 54.5%).
Dolphin Research’s take: MI355 under-delivers; leaning on rack-scale MI450
Results were decent, with revenue and margins meeting or beating expectations, and growth driven by a CPU rebound.
But the market is disappointed with MI355’s QoQ performance. Since most of the Data Center delta came from server CPUs, Dolphin Research estimates AI GPU revenue at ~$2.4bn this quarter.
Ex the $390mn contribution from MI308 to China, AI GPUs (incl. MI355) were only ~$2.0bn, up just ~$150mn QoQ. MI355 only began mass production in 2H25, yet QoQ momentum already slowed into Q4.
With Q1 guidance implying ~5% QoQ revenue decline, and seasonal softness in Client, Gaming, and Embedded, Dolphin Research estimates AI GPU could slip to ~$2.2bn next quarter. Ex the ~$100mn MI308 contribution, the rest of AI GPU may be up only ~$100mn QoQ.
Management cited customer readiness, noting roughly $1.5bn of Q1 orders would shift into Q2, but two straight quarters of soft MI355 ramps will prompt questions on AI GPU product strength so soon after mass production began.
Beyond this print, the market is watching three areas:
a) Hyperscaler capex: In this hot AI cycle, hyperscalers remain the key chip buyers.
Meta and Microsoft reported first. Meta raised 2026 capex to $115–135bn, +60% to +87% YoY, well above prior expectations (~$110bn).
For 2026, Dolphin Research expects combined capex at Meta, Google, Microsoft, and Amazon to exceed $600bn, up ~50% YoY, implying strong support for server and AI silicon demand.
By cadence, 2026 should be back-half weighted, with NVIDIA’s Rubin and AMD’s MI400 families ramping in 2H.
b) CPU competitiveness: AMD keeps chipping away at Intel’s CPU share.
With PCs up 11% YoY this quarter, Intel’s Client revenue still fell 6.6% YoY, while AMD grew 34%.
This divergence underscores AMD’s rising CPU edge. Across total CPUs (PC + server), AMD’s share continues to climb, and has overtaken Intel in desktops.
c) AI GPU roadmap: Today’s shipment workhorse is MI355, with next-gen MI455 due in 2H.
Upgrades from MI355 to MI455 focus on compute, memory, and clustering. Compute is expected to roughly double (moving to 2nm), while memory moves from HBM3E to HBM4, boosting bandwidth and capacity.
More than raw compute or memory, the market cares about breakthroughs in rack-scale solutions. As workloads shift from training to inference, compute matters less. For example, MI355’s compute peers NVIDIA’s B200 tier, but AMD’s key gap is the lack of rack-scale cluster delivery.
For the MI450 family, AMD plans to tightly couple it with the Helios rack platform to deliver rack-scale clusters rather than single chips, directly targeting NVIDIA’s Rubin architecture.
The Helios rack supports up to 72 MI455X GPUs, enabling cross-GPU and cross-rack coordination via 43TB/s east-west bandwidth (ultra-ethernet) and full liquid cooling, delivering 2.9 ExaFLOPS (FP4) per rack.
Hyperscalers need cluster-scale deployments, not just boards. MI450’s rack-scale delivery aligns with CSP demand, but with volume only in 2H, AI revenue in 2026 should be back-half weighted.
As MI450 ramps, Data Center growth could re-accelerate.
At the current market cap of $394.2bn, AMD trades at roughly 38x 2026 core net profit (assumes revenue +44% YoY, GPM 53.9%, tax rate 13%). Versus AI peers, AMD’s PE screens above NVIDIA (23x) and TSMC (22x), reflecting expectations for share gains and high growth in AI.
Near term, results hinge on CPUs and MI355 shipments; a softer MI355 QoQ likely pressures the stock. Medium to long term, focus shifts to MI450 progress: new customers, orders, and rack-scale cluster delivery.
With AI shifting toward inference, compute intensity matters less while CPU demand rises, which benefits AMD. As CPU demand recovers, AMD’s base earnings should grind higher.
AMD still trails NVIDIA on peak compute, but that does not preclude inference wins. Even if MI355 underwhelms now, the MI450 rack-scale solution due in 2H could be a catalyst after any pullback.
Below is Dolphin Research’s detailed read of AMD’s print:
I. Headline: MI308 inventory release supports continued growth
1.1 Revenue
Q4 2025 revenue was $10.27bn, +34% YoY, beating the raised consensus (~$10.0bn). The YoY increase was mainly driven by data center CPUs and Client.
Data Center rose by roughly $1.0bn QoQ, but ~$390mn came from MI308 sales in China. Ex this, AI GPU revenue rose only ~$150mn QoQ, with most of the incremental contribution from server CPUs.
1.2 Gross profit</strong>
Gross profit was $5.58bn, +44% YoY, with GPM at 54.3%, a clear rebound.
GPM had previously dipped to 39.8% due to MI308 inventory impairment (ex-impairment ~50.2%). This quarter benefited from MI308 sell-through (~$360mn); ex this, underlying GPM was ~50.8%, roughly flat YoY.
1.3 Opex
Total opex was $3.53bn, +41% YoY, with both R&D and SG&A increasing.
Specifically: (i) R&D was $2.33bn (+36% YoY), with ongoing AI-focused investments; (ii) SG&A was $1.20bn (+51% YoY), tracking revenue and including higher employee incentives.
1.4 Profit
Amortization related to the Xilinx acquisition will weigh on reported profit for some time, so core operating profit offers a cleaner view.
Core OP = GP − R&D − SG&A.
Ex acquisition-related items, Dolphin Research estimates Q4 core OP at ~$2.05bn, continuing to recover. Even after stripping the ~$360mn MI308 margin benefit, core OP still improved, supported by the server CPU rebound.
II. Segments: MI355 QoQ slows; MI455 is the key swing factor
Data Center and Client remain the core businesses, together over 80% of revenue. With higher AI GPU shipments and a server CPU rebound, Data Center mix continued to rise.
2.1 Data Center
Q4 Data Center revenue was $5.38bn, +39% YoY, in line with the higher consensus (~$5.3bn). Growth was driven by server CPUs and renewed MI308 sales.
Based on company and market inputs, Dolphin Research estimates AI GPU at ~$2.4bn and data center CPU/other at ~$2.98bn this quarter.
With $390mn from MI308, the rest of AI GPUs (incl. MI355) were ~ $2.0bn, up only ~$150mn QoQ, short of market hopes.
Specifically:
a) Data center CPU: still strong. With Intel’s CPU lineup lagging, AMD’s combined CPU+GPU offering is driving server CPU share above 20%. Despite Intel’s x86 collaboration with NVIDIA, AMD’s MI roadmap should continue to pull server CPU demand.
b) AI GPU: soft again. The main delta this quarter was MI308 sell-through; the rest of AI GPUs (incl. MI355) rose only ~$150mn QoQ. Given MI355 only ramped in 2H25, that QoQ outcome is underwhelming.
For next quarter, Dolphin Research expects AI GPU revenue at ~$2.2bn, incl. ~$100mn from MI308, implying MI355 may add only ~$100mn QoQ.
With Meta and other hyperscalers lifting capex, Dolphin Research sees combined 2026 capex for Meta, Google, Microsoft, and Amazon exceeding $600bn (+50% YoY), underpinning robust AI chip growth in 2026.
In such a high-growth market, AMD’s upside depends on product strength. In server CPUs, AMD is still taking share; in AI GPUs, MI355 has disappointed, pushing attention to the MI450 family.
AMD plans to bind MI450 tightly to the Helios rack platform to move from single-die to rack-scale cluster delivery, directly targeting NVIDIA’s Rubin and aligning with CSP needs.
2.2 Client
Client revenue was $3.1bn, +34% YoY, beating consensus (~$2.9bn). Growth was driven by PC market recovery and share gains vs. Intel.
Industry data show Q4 2025 global PC shipments at 76.4mn units, +11% YoY. By contrast, AMD’s Client revenue grew 34% YoY, while Intel’s Client revenue fell 6.6% YoY.
The spread in growth suggests a continued PC recovery with AMD gaining share, including an overtaking of Intel in desktops.
2.3 Other
1) Gaming: Q4 Gaming revenue was $840mn, +49% YoY.
Specifically, (i) Semi-custom: up YoY but down QoQ; 2026 enters year 7 of the console cycle, with the next Xbox (AMD semi-custom SoC) expected in 2027. (ii) Gaming GPU: strong holiday demand for Radeon RX 9000s boosted channel sell-through.
2) Embedded: Q4 Embedded revenue was $950mn, +3% YoY, supported by demand recovery across several end markets.
<End>
Related AMD research from Dolphin Research
AMD earnings season
Nov 5, 2025 call Trans: AMD (Trans): No new orders disclosed; MI450 focuses on rack-scale solutions
Nov 5, 2025 earnings First Take: AMD: Tied to OpenAI? Not a 'backup' get-out-of-jail-free card
Aug 6, 2025 call Trans: AMD (Trans): AI GPU to grow YoY next quarter even excluding MI308
Aug 6, 2025 earnings First Take: AMD: CPUs outgunning Intel; when can AI GPUs challenge NVIDIA?
May 7, 2025 call Trans: AMD (Trans): AI GPU to deliver double-digit growth for the year
May 7, 2025 earnings First Take: AMD: Dominating Intel, turning up the heat on NVIDIA?
Feb 5, 2025 call Trans: AMD (Trans): No sequential growth in Data Center in 1H25
Feb 5, 2025 earnings First Take: AMD: Deepseek pours fuel; the GPU 'backup' dream hits ASIC reality
Risk disclosure and statement: Dolphin Research disclaimer and general disclosures
The copyright of this article belongs to the original author/organization.
The views expressed herein are solely those of the author and do not reflect the stance of the platform. The content is intended for investment reference purposes only and shall not be considered as investment advice. Please contact us if you have any questions or suggestions regarding the content services provided by the platform.
